TheAnachronism/docspell
1
0
Fork 0
mirror of https://github.com/TheAnachronism/docspell.git synced 2025-03-28 09:45:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

Update documentation

Browse Source
This commit is contained in:
Eike Kettner 2020-09-02 00:18:55 +02:00
parent 4309bd8dfd
commit 145c308461
2 changed files with 21 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
website/elm/Feature.elm
View File

@ -67,7 +67,7 @@ Text is extracted from all files. For scanned documents/images, OCR is used by u
, { image = "img/analyze-feature.png"
, header = "Text Analysis"
, description = """
The extracted text is analyzed and is used to find properties that can be annotated to your documents automatically.
The extracted text is analyzed using ML techniques to find properties that can be annotated to your documents automatically.
"""
}
, { image = "img/filetype-feature.svg"

25
website/site/content/docs/webapp/metadata.md
View File

@ -33,11 +33,26 @@ workflows, a tag category *state* may exist that includes tags like
"assignment" semantics. Docspell doesn't propose any workflow, but it
can help to implement some.
The tags are *not* taken into account when creating suggestions from
analyzed text yet. However, PDF files may contain metadata itself and
if there is a metadata *keywords* list, these keywords are matched
against the tags in the database. If they match, the item is tagged
automatically.
Docspell can try to predict a tag for new incoming documents
automatically based on your existing data. This requires to train an
algorithm. There are some caveats: the more data you have correctly
tagged, the better are the results. So it won't work well for maybe
the first 100 documents. Then the tags must somehow relate to a
pattern in the document text. Tags like *todo* or *waiting* probably
won't work, obviously. But the typical "document type" tag, like
*invoice* and *receipt* is a good fit! That is why you need to provide
a tag category so only sensible tags are being learned. The algorithm
goes through all your items and learns patterns in the text that
relate to the given tags. This training step can be run periodically,
as specified in your collective settings such that docspell keeps
learning from your already tagged data! More information about the
algorithm can be found in the config, where it is possible to
fine-tune this process.
Another way to have items tagged automatically is when an input PDF
file contains a list of keywords in its metadata section (this only
applies to PDF files). These keywords are then matched against the
tags in the database. If they match, the item is tagged with them.
## Organization and Person